The operational ALADIN models

1. Status report of the operational ALADIN versions

ALADIN runs operationally on a wide range of computers, from single workstations to vectorial computers in shared or distributed memory, and on a cluster of workstations under Linux. 12 operational ALADIN suites are presently running in various NMS. 7 coupling domains are produced operationally at Météo-France, 1 in Prague.

See updated informations on this web server at : http://www.cnrm.meteo.fr/aladin/oper/oper.html .

2. Operational version at Austrian Meteorological Service

Last news in Newsletter16. 

3. Operational implementation of ALADIN-Belgium

No change since 16/11/1999 (introduction of CYCORA scheme) :

• operational version : AL11T2_export.02
• 48h forecast, twice a day,
• coupling every 3 hours from Aladin-France.
• basic output every hour, wider output every 3 hours
• running in shared memory mode on the Cray J90, with 8 processors
• successfull tests of MPI on the J90.

The tests on HP machines meet difficulties due to unsatisfying coding, that defines specific kinds (JPRB for instance) but which actually are not fully supported.

System libraries and some xrd routines often expect default kinds, while ALADIN passes the user-defined kinds.

This problem is often solved by compiling on workstations with "+autodbl"-like flags, but on HP those flags are not allowed when compiling for MPI.

The replacement of the J90 is planned by another system, shared with the Royal Observatory and the Space Aeronomy Institute, but the choice has not yet been finalized.

4. Operaional version at Bulgarian Meteorological Service

Last news in Newsletter16.

5. Operational ALADIN-FRANCE in Météo-France

Since the 20th of june at 12H00 UTC, ALADIN coupling files are produced by an ARPEGE based on a variational analysis with four dimensions (4DVAR). 4DVAR consists in minimising the distance between the model trajectory and observations over a temporal range (currently 6 hours).

The tests done showed substancial improvements at the global scale and particularly over the southern hemisphere.These changes are due to :

• a better use of asynoptic observations as satellites, planes data and hourly synop which are considered +/- half an hour around their observation time ;
• increments produced by an observation are coherent with the meteorological situation : an information about baroclinic instability will modify the first guess in a baroclinic way ;
• the model is used in the analysis to get the trajectories so that the "shocks" between the first guess and the analysis are reduced.

6. Workstation version at French Meteorological Service

No differences now between the operational ALADIN-France version and the workstation version.

On the workstation SUN, with the compilator F90 Fujitsu and the ALADIN version export_AL11T2_01 (with some bugfix), tests were executed with the option LMESSP=TRUE (mono processor and distributed memory). Using the two methods (dummy message passing and modifications of Josef Vivoda and Andrey Bogatchev on the MPI routines), the Fpos configuration can run but within an abnormally long time, of course for a big domain as FRANX01 (about 2221281 points).

Also a version AL13 (al13_main.01, cy22t2_main.01 and xr22_main.02) is going to be implemented. Some tests, for the configurations 001 and Fpos, have been made on the French domain, with the option LMESSP to FALSE (shared memory). For instance, the use of the tool progrid shows that the combination (Fpos and progrid) produces the same grib files as the previous cycle. As the new source contains the modifications of Josef Vivoda and Eric Sevault, some tests with the option LMESSP= TRUE were made, but unfortunately the large drift (about 3 times more) in the performance remains. Further investigations are required to understand and reduce this extra-cost.

7. Operational version at Hungarian Meteorological Service

Regarding the operational setting there were no major changes encountered during the first quarter of 2000: two model runs/day and 24 diag.pack (CANARI) runs/day are the main operational jobs.

At the begining of the year a new binary was created for the ALADIN code usingthe native mpi library of the SGI machine (instead of MPICH).

The model integration is now using 10 processors and the two remaining processors are devoted to the auxilary jobs (like ee927, file preparations, etc.).

Work had been started to port the FESTATt tool (computation of background error statistics) to the HMS environment. The first preliminary statistics and plots were obtained for the end of the period.

The generation of automatic forecasts was extended over 65 locations using the raw model products of the ALADIN/HU model.

During the second quarter of 2000, there were two main changes in the operational suite :

• On the request of the aviation meteorologists we have extended the hourly post-processing frequency until 24 hours (so during the 48 hours forecast post-processed outputs are produced hourly until 24 hours and 3 hourly afterwards).
• Cycle AL12 was installed and put into operations on the last days of June.

Other important events of interest:

• A two-weeks long workshop took place in Budapest with the primary aim to run ALADIN 3DVAR on the Origin 2000 machine in Budapest. This goal was achieved and now we are able to run 3DVAR ALADIN on the 12 processors of our computer (a detailed report is available in this Newsletter).
• AL11T2 was also installed with the aim to run the non-hydrostatic version of the ALADIN model. There are still some problems (instabilities) detected, which are going to be investigated in the near future.
• An action related to the surface parametrisation of the model was continued and a local soil texture data base was included into the climatological information of the ALADIN model. The effect of the more detailed and realistic data over Hungary will be studied in the near future.

8. Operational ALADIN-LACE in CHMI

1. Evolution of the ALADIN/LACE application

To start with, the entering of the last year of the 20th century went rather well, without any real trouble caused by the famous Y2K bug. Only a small mistake (in fact two mistakes compensating each other) was detected later on in the encoding of date to GRIB bulletins and corrected. Also on the 29th February 2000 everything went smoothly….

Following the positive scores of the parallel suite named "HUCOE/HUTIL", the change was introduced into the operational ALADIN/LACE application on 23/02/2000 for 12 UTC network time.

12/04/2000 for 12 UTC network time: "quadratic coupling": The "quadratic coupling" means that the time interpolation of the large-scale values between the boundary condition refreshments is parabolic in time and not any more linear. Like that the LBC trajectory is described with a higher precision and leads to a small improvement in the general scores. This technique is, however, also quite sensible to be used in rapidly evolving situations, like it was the case of the famous Christmas'99 storm, see some details below in this report. The quadratic coupling belonging to the improvements first applied in ALADIN/LACE and yet nowhere else, further details may be found at the usual web address of RC LACE, the parallel suite had name "aav".

2. Parallel Suites

During the first quarter of 2000, the Prague Team launched the following parallel tests to assess the impact of different modifications:

• HUCOE/HUTIL suites, modifying the parameterization of clouds (threshold humidity values to have a cloud or not, including parameterization of the depth of the cloud; suites aar, aat). The suite "aat" entered the operational application, with new values of HUCOE=2., HUTIL=1.8.
• 41Levels suite "aas", testing the impact of higher vertical resolution (anticipation of ARPEGE 41L configuration). The suite revealed a problem of noise due to extrapolation of wind to the stratospheric extra-levels. The scores were mostly negative. It lead to a re-design of future spacing of vertical levels, see R & D.
• Non-linear horizontal diffusion made with Semi-Lagrangian interpolators research suites (aaw, aax). These suites were meant to assess the impact caused by the new way of applying the horizontal diffusion. It helpfully revealed the pluses and minuses of the new scheme. For details see the previous reports on PhD work of Filip Vana.
• LQCPL suite (aav), testing the quadratic interpolation of the lateral boundary conditions in time. The results are slightly positive in geopotential scores in the troposphere.
• Cycle AL12 verification suite (aay), confirming the neutral results with respect to the currently operational AL11/CY21T1+operational fix CYCORA.

The scores may be consulted on www.chmi.cz/meteo/ov/lace/aladin/partests pages.

Otherwise the cycle AL12/CY22/XR22 was successfully ported to SX4 platform and validated. A bug in the cray-type of multitasking (key LMLTSK) has been detected and corrected. The bug concerns other ALADIN teams using Cray platforms. More details on AL12 can be asked to: Gabor Radnoti.

During the second quarter of 2000, the Prague Team launched the following parallel tests to assess the impact of different modifications:

• 37 Levels suite (aaz), running for about two month. The scores are either neutral or positive in the troposphere. However, there is worse RMSE score of geopotential in the stratosphere, limited, it seems, to several days in April. Further investigation of this score is still necessary. For the time being the suite aaz was stopped.
• Blending suite, still running and subjected to careful validations, scheduled for the month of July.
• Linear grid suite (aba) has been started at the end of June. It consists in using a linear grid type of spectral truncation for all variables but orography. The orography remains the same like it is when using the quadratic grid, in order to avoid 2 delta x waves amplitudes to get excited because of the orographic forcing. This suite reflects recent results brought by the quasi-academic experiment SCANIA.

The scores may also be consulted on www.chmi.cz/meteo/ov/lace/aladin/partests pages.

9. Operational ALADIN-MAROC in MAROC-Météo

(more details Radi Ajjaji)

1. Operational suite :

ALADIN-Maroc is running with ISBA assimilation cycle twice a day with 16,5 km resolution, 675s time step and 31 levels. The model version is based on cycle AL08.

2. Operational development :

• Improvement of operational suite script's in order to disable the read errors.
• Test of an Aladin workstation version over Morocco.
• Elaboration of a pre-operational statistical adaptation suite of the maximal and the minimal temperature.

3. Computer facilities:

Maroc-Meteo is planning nearly to upgrade the resolution of the Aladin-Maroc model. For this, a call for proposal is done in order to buy a new powerful computer.

10. Operational version at Polish Meteorological Service

Last news in Newsletter16. 

11. Operational version at the Portuguese Meteorological Service

(more details mario.almeida++at++meteo.pt )

ALADIN/Portugal was declared operational on 24th of April. The AL11T2 cycle was installed in the begin of the year and tested in parallel suite during 2 months (April and May). This cycle has replaced the AL09 one in the operational suite, on June 27 th , 12 UTC run, and the frequency of model outputs was switched from 3h to 1h in operational configuration. During June the new model version -AL12- was successful implemented.

The team reinforcement has allowed us to restart the work to create the CMAFOC files for OI/CANARI in order to process locally the TEMP and surface observations.

12. Operational version at the Romanian Meteorological Service)

From first of June 2000 the operational version of the Aladin model in Bucharest is al12. The option LMESSP=.F. is still used due to our workstation limitation.

Other changes of the operational systems concern:

• The possibility to use the forecast vertical profiles (emagrammes) under the visualisation MESSIR system.
• The statistical adaptation: the correction of the 2m temperature using Kalman filter is computed for 140 meteorological stations. Daily the verification is carried out also for the corrected temperatures in comparison with the direct model output. Here is an example of the monthly statistical measures for 2000, June ( mean error, mean absolute error, root mean square error).
• The Black Sea wave prediction system was developed by a nested integration over the western basin (for a grid of 5' x 5') allowing the use of the high-resolution wind Aladin forecast every 3 hours.
• The forecast of a thermal comfort index based on the Aladin output.

13. Operationa version at Slovak Meteorological Service

Last news in Newsletter16. 

14. Operational version at Slovenian Meteorological Service

Last news in Newsletter16 .

Home